The (synchronous) 0.4s per move number is misleading (and wrong), that's not what the paper is saying. The "footnote 1" of the article is wrong.